Prediction of Freezing Time During Hydrogen Fueling Using Machine Learning

This study presents a method for predicting nozzle surface temperature and the timing of frost formation during hydrogen refueling using machine learning. A continuous refueling system was implemented based on a simulation model that was developed and validated in previous research. Data were collected under various boundary conditions, and eight regression models were trained and evaluated for their predictive performance. Hyperparameter optimization was performed using random search to enhance model performance. The final models were validated by applying boundary conditions not used during model development and comparing the predicted values with simulation results. The comparison revealed that the maximum error rate occurred after the second refueling, with a value of approximately 4.79%. Currently, nitrogen and heating air are used for defrosting and frost reduction, which can be costly. The developed machine learning models are expected to enable prediction of both frost formation and defrosting timings, potentially allowing for more cost-effective management of defrosting and frost reduction strategies.